diff --git a/MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp b/MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp
index 4a8b9238..05447c79 100644
--- a/MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp
+++ b/MobileGL/MG_Backend/DirectGLES/DirectGLES.cpp
@@ -22,6 +22,9 @@
 #include <MG_Util/Converters/MGToStr/TextureEnumConverter.h>
 #include <MG_Util/Converters/MGToGL/RenderStateEnumConverter.h>
 #include <MG_Util/Texture/PixelStoreProcessor.h>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
 #if defined(__linux__) && !defined(__ANDROID__) && __has_include(<X11/Xlib.h>)
 #pragma push_macro("Bool")
 #pragma push_macro("None")
@@ -35,6 +38,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
     MG_External::GLESFunctionsTable g_GLESFuncs;
     MG_External::GLESCapabilities g_GLESCapabilities;
 
+    static Bool QueryCurrentSurfaceSize(Int& outWidth, Int& outHeight);
+
     enum class DrawSyncBit : Uint32 {
         None = 0,
         IndexBuffer = 1 << 0,
@@ -375,19 +380,30 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     namespace RenderStateImpl {
         static Uint16 g_syncedRenderStateVersion = 0;
+        static Bool g_hasSyncedRenderState = false;
         static RenderStateParameters g_syncedRenderStateParameters;
+        static IntVec4 g_syncedBackendViewport = IntVec4(-1, -1, -1, -1);
         void SyncRenderState() {
 #ifdef TRACY_ENABLE
             ZoneScopedC(TRACY_ZONECOLOR_BACKEND);
 #endif
             Uint16 currentRenderStateVersion = MG_State::pGLContext->GetRenderStateParametersVersion();
-            if (currentRenderStateVersion == g_syncedRenderStateVersion) return;
+            if (g_hasSyncedRenderState && currentRenderStateVersion == g_syncedRenderStateVersion) return;
 
             const auto& parameters = MG_State::pGLContext->GetRenderStateParameters();
 
-            if (parameters.Viewport != g_syncedRenderStateParameters.Viewport) {
-                g_GLESFuncs.glViewport(parameters.Viewport.x(), parameters.Viewport.y(), parameters.Viewport.z(),
-                                       parameters.Viewport.w());
+            IntVec4 backendViewport = parameters.Viewport;
+            if (backendViewport.z() <= 0 || backendViewport.w() <= 0) {
+                Int surfaceWidth = 0;
+                Int surfaceHeight = 0;
+                if (QueryCurrentSurfaceSize(surfaceWidth, surfaceHeight)) {
+                    backendViewport = IntVec4(0, 0, surfaceWidth, surfaceHeight);
+                }
+            }
+            if (backendViewport != g_syncedBackendViewport) {
+                g_GLESFuncs.glViewport(backendViewport.x(), backendViewport.y(), backendViewport.z(),
+                                       backendViewport.w());
+                g_syncedBackendViewport = backendViewport;
             }
 
 #define SYNC_CAPABILITY(cap_mg, cap_gl)                                                                                \
@@ -552,6 +568,7 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
             g_syncedRenderStateVersion = currentRenderStateVersion;
             g_syncedRenderStateParameters = parameters;
+            g_hasSyncedRenderState = true;
         }
     } // namespace RenderStateImpl
 
@@ -1873,6 +1890,73 @@ namespace MobileGL::MG_Backend::DirectGLES {
     static EGLSurface g_Surface = EGL_NO_SURFACE;
     static EGLConfig g_Config = nullptr;
 
+    static Bool QueryCurrentSurfaceSize(Int& outWidth, Int& outHeight) {
+        outWidth = 0;
+        outHeight = 0;
+        if (!g_EGLFuncs.eglQuerySurface || g_Display == EGL_NO_DISPLAY || g_Surface == EGL_NO_SURFACE) {
+            return false;
+        }
+
+        EGLint width = 0;
+        EGLint height = 0;
+        if (!g_EGLFuncs.eglQuerySurface(g_Display, g_Surface, EGL_WIDTH, &width) ||
+            !g_EGLFuncs.eglQuerySurface(g_Display, g_Surface, EGL_HEIGHT, &height) ||
+            width <= 0 || height <= 0) {
+            return false;
+        }
+
+        outWidth = static_cast<Int>(width);
+        outHeight = static_cast<Int>(height);
+        return true;
+    }
+
+    static Bool PresentStatsEnabled() {
+        static const Bool enabled = [] {
+            const char* value = std::getenv("MOBILEGL_GLES_PRESENT_STATS");
+            return value != nullptr && value[0] != '\0' && std::strcmp(value, "0") != 0;
+        }();
+        return enabled;
+    }
+
+    static void DumpDefaultFramebufferStats() {
+        if (!PresentStatsEnabled() || !g_GLESFuncs.glReadPixels || !g_EGLFuncs.eglQuerySurface ||
+            g_Display == EGL_NO_DISPLAY || g_Surface == EGL_NO_SURFACE) {
+            return;
+        }
+
+        Int width = 0;
+        Int height = 0;
+        if (!QueryCurrentSurfaceSize(width, height)) {
+            return;
+        }
+
+        GLint viewport[4] = {0, 0, 0, 0};
+        g_GLESFuncs.glGetIntegerv(GL_VIEWPORT, viewport);
+        GLint previousReadFramebuffer = 0;
+        g_GLESFuncs.glGetIntegerv(GL_READ_FRAMEBUFFER_BINDING, &previousReadFramebuffer);
+        g_GLESFuncs.glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
+
+        Vector<Uint8> pixels(static_cast<SizeT>(width) * static_cast<SizeT>(height) * 4);
+        g_GLESFuncs.glReadPixels(0, 0, width, height, GL_RGBA, GL_UNSIGNED_BYTE, pixels.data());
+
+        SizeT nonBlack = 0;
+        SizeT nonZeroAlpha = 0;
+        for (SizeT offset = 0; offset + 3 < pixels.size(); offset += 4) {
+            if (pixels[offset] != 0 || pixels[offset + 1] != 0 || pixels[offset + 2] != 0) {
+                ++nonBlack;
+            }
+            if (pixels[offset + 3] != 0) {
+                ++nonZeroAlpha;
+            }
+        }
+
+        g_GLESFuncs.glBindFramebuffer(GL_READ_FRAMEBUFFER, static_cast<GLuint>(previousReadFramebuffer));
+        std::fprintf(stderr,
+                     "MOBILEGL_GLES_PRESENT_STATS nonBlack=%zu/%zu alpha=%zu/%zu size=%dx%d viewport=%d,%d,%d,%d\n",
+                     nonBlack, pixels.size() / 4, nonZeroAlpha, pixels.size() / 4, width, height,
+                     viewport[0], viewport[1], viewport[2], viewport[3]);
+    }
+
 #if defined(__linux__) && !defined(__ANDROID__)
     static void* OpenX11Lib() {
         void* x11Lib = dlopen("libX11.so.6", RTLD_LOCAL | RTLD_NOW);
@@ -2092,6 +2176,10 @@ namespace MobileGL::MG_Backend::DirectGLES {
 
     void Present() {
         if (g_Display != EGL_NO_DISPLAY && g_Surface != EGL_NO_SURFACE) {
+            DumpDefaultFramebufferStats();
+            if (g_GLESFuncs.glFlush) {
+                g_GLESFuncs.glFlush();
+            }
             g_EGLFuncs.eglSwapBuffers(g_Display, g_Surface);
         }
     }
diff --git a/MobileGL/MG_Backend/DirectGLES/Managers.cpp b/MobileGL/MG_Backend/DirectGLES/Managers.cpp
index bb284c22..144aca17 100644
--- a/MobileGL/MG_Backend/DirectGLES/Managers.cpp
+++ b/MobileGL/MG_Backend/DirectGLES/Managers.cpp
@@ -27,6 +27,17 @@
 namespace MobileGL::MG_Backend::DirectGLES {
     constexpr Bool PREFER_MAP_BUFFER_RANGE_FOR_BUFFER_SYNC = true;
 
+    static Uint ResolveBackendEsslVersion() {
+        const auto& version = g_GLESCapabilities.GLESVersion;
+        if (version.Major > 3 || (version.Major == 3 && version.Minor >= 2)) {
+            return 320;
+        }
+        if (version.Major == 3 && version.Minor >= 1) {
+            return 310;
+        }
+        return 300;
+    }
+
     namespace BufferImpl {
         BackendBufferObject::BackendBufferObject() {
 #ifdef TRACY_ENABLE
@@ -1181,8 +1192,8 @@ namespace MobileGL::MG_Backend::DirectGLES {
                 spvc_compiler_options options;
                 spvcSession.CreateOptions(&options);
 
-                // TODO: check ESSL version supported by backend driver
-                spvc_compiler_options_set_uint(options, SPVC_COMPILER_OPTION_GLSL_VERSION, 320);
+                spvc_compiler_options_set_uint(options, SPVC_COMPILER_OPTION_GLSL_VERSION,
+                                               ResolveBackendEsslVersion());
                 spvc_compiler_options_set_bool(options, SPVC_COMPILER_OPTION_GLSL_ES, SPVC_TRUE);
                 spvc_compiler_options_set_bool(options, SPVC_COMPILER_OPTION_GLSL_VULKAN_SEMANTICS, SPVC_FALSE);
 
diff --git a/MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp b/MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp
index 7210626a..564c9aee 100644
--- a/MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp
+++ b/MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp
@@ -1634,15 +1634,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
 
     const ProgramFactory::VkProgramObject& ProgramFactory::GetOrCreateProgram(
         const MG_State::GLState::ProgramObject& program, CompileOptionFlags flags) {
-        const Uint32 backendStateVersion = program.GetBackendStateVersion();
-        HashType hash = 0;
-        if (m_lastLookup.program == &program && m_lastLookup.backendStateVersion == backendStateVersion &&
-            m_lastLookup.flags == flags) {
-            hash = m_lastLookup.hash;
-        } else {
-            hash = ComputeHash(program, flags);
-            m_lastLookup = {.program = &program, .backendStateVersion = backendStateVersion, .flags = flags, .hash = hash};
-        }
+        const HashType hash = ComputeHash(program, flags);
         auto it = m_cache.find(hash);
         if (it != m_cache.end()) {
             return it->second;
diff --git a/MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp b/MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp
index fef6a2ba..0a39abdc 100644
--- a/MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp
+++ b/MobileGL/MG_Backend/DirectVulkan/Renderer/UniformManager.cpp
@@ -16,6 +16,9 @@
 #include "MG_Util/Converters/MGToStr/FramebufferEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
 #include "MG_Util/Metrics/TextureMetrics.h"
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
 #include <limits>
 
 namespace MobileGL::MG_Backend::DirectVulkan {
@@ -74,6 +77,14 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return uniformUnit >= 0 ? uniformUnit : 0;
     }
 
+    static Bool ShouldDumpDescriptorStats() {
+        static const Bool enabled = [] {
+            const char* value = std::getenv("MOBILEGL_DESCRIPTOR_STATS");
+            return value != nullptr && value[0] != '\0' && std::strcmp(value, "0") != 0;
+        }();
+        return enabled;
+    }
+
     Bool UniformManager::Initialize(VkDevice device, VkBufferManager* bufferManager,
                                              ProgramFactory* programFactory,
                                              VkDeviceSize minUniformBufferOffsetAlignment, Uint32 frameCount,
@@ -262,6 +273,21 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             .imageView = resource->fullView,
             .imageLayout = resource->layout,
         };
+        if (ShouldDumpDescriptorStats()) {
+            std::fprintf(stderr,
+                         "MOBILEGL_DESCRIPTOR_STATS program=%u binding=%u name=%s location=%d unit=%d "
+                         "texture=%u target=%d sampler=%p imageView=%p layout=%d\n",
+                         program.GetExternalIndex(),
+                         binding,
+                         programObj.samplerNameByBinding[binding].c_str(),
+                         location,
+                         unit,
+                         texture->GetExternalIndex(),
+                         static_cast<Int>(texture->GetTarget()),
+                         reinterpret_cast<void*>(outImageInfo.sampler),
+                         reinterpret_cast<void*>(outImageInfo.imageView),
+                         static_cast<Int>(outImageInfo.imageLayout));
+        }
         return outImageInfo.sampler != VK_NULL_HANDLE;
     }
 
diff --git a/MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.cpp b/MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.cpp
index 382d4519..d48ba0fe 100644
--- a/MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.cpp
+++ b/MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.cpp
@@ -12,6 +12,9 @@
 #include "MG_Util/Converters/MGToStr/TextureEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
 
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
 #include <memory>
 
 namespace MobileGL::MG_Backend::DirectVulkan {
@@ -316,6 +319,99 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return true;
     }
 
+    static Bool ShouldDumpTextureUploadStats() {
+        static const Bool enabled = [] {
+            const char* value = std::getenv("MOBILEGL_TEXTURE_UPLOAD_STATS");
+            return value != nullptr && value[0] != '\0' && std::strcmp(value, "0") != 0;
+        }();
+        return enabled;
+    }
+
+    static void DumpTextureSyncStats(Int textureId, TextureInternalFormat format, TextureUploadTarget uploadTarget,
+                                     Uint32 mipLevelCount, const IntVec3& texelSize, SizeT byteSize,
+                                     Bool hasDirtyMipLevel) {
+        if (!ShouldDumpTextureUploadStats()) {
+            return;
+        }
+
+        std::fprintf(stderr,
+                     "MOBILEGL_TEXTURE_SYNC_STATS texture=%d format=%s target=%s mips=%u size=%dx%dx%d "
+                     "bytes=%zu dirty=%d\n",
+                     textureId,
+                     MG_Util::ConvertTextureInternalFormatToString(format).c_str(),
+                     MG_Util::ConvertTextureUploadTargetToString(uploadTarget).c_str(),
+                     mipLevelCount,
+                     texelSize.x(),
+                     texelSize.y(),
+                     texelSize.z(),
+                     byteSize,
+                     hasDirtyMipLevel ? 1 : 0);
+    }
+
+    static void DumpTextureUploadStats(Int textureId, TextureUploadTarget target, Uint32 level,
+                                       const IntVec3& texelSize, const void* data, SizeT byteSize, Uint32 channels) {
+        if (!ShouldDumpTextureUploadStats() || data == nullptr || byteSize == 0 || channels == 0) {
+            return;
+        }
+
+        const auto* bytes = static_cast<const Uint8*>(data);
+        Uint8 minValue = 255;
+        Uint8 maxValue = 0;
+        SizeT nonZero = 0;
+        Uint64 sum = 0;
+        Uint64 neighborDiff = 0;
+        SizeT neighborCount = 0;
+
+        for (SizeT i = 0; i < byteSize; ++i) {
+            minValue = std::min(minValue, bytes[i]);
+            maxValue = std::max(maxValue, bytes[i]);
+            nonZero += bytes[i] != 0 ? 1 : 0;
+            sum += bytes[i];
+        }
+
+        const SizeT width = static_cast<SizeT>(std::max(texelSize.x(), 0));
+        const SizeT height = static_cast<SizeT>(std::max(texelSize.y(), 0));
+        const SizeT pixelStride = channels;
+        if (width > 1 && height > 0 && byteSize >= width * height * pixelStride) {
+            for (SizeT y = 0; y < height; ++y) {
+                const SizeT row = y * width * pixelStride;
+                for (SizeT x = 1; x < width; ++x) {
+                    const SizeT prev = row + (x - 1) * pixelStride;
+                    const SizeT cur = row + x * pixelStride;
+                    for (SizeT c = 0; c < std::min<SizeT>(channels, 3); ++c) {
+                        neighborDiff += static_cast<Uint64>(
+                            std::abs(static_cast<Int>(bytes[cur + c]) - static_cast<Int>(bytes[prev + c])));
+                        ++neighborCount;
+                    }
+                }
+            }
+        }
+
+        const double avg = byteSize > 0 ? static_cast<double>(sum) / static_cast<double>(byteSize) : 0.0;
+        const double avgNeighborDiff =
+            neighborCount > 0 ? static_cast<double>(neighborDiff) / static_cast<double>(neighborCount) : 0.0;
+        std::fprintf(stderr,
+                     "MOBILEGL_TEXTURE_UPLOAD_STATS texture=%d target=%s level=%u size=%dx%dx%d bytes=%zu "
+                     "channels=%u min=%u max=%u avg=%.2f nonzero=%zu neighborDiff=%.2f first=%u,%u,%u,%u\n",
+                     textureId,
+                     MG_Util::ConvertTextureUploadTargetToString(target).c_str(),
+                     level,
+                     texelSize.x(),
+                     texelSize.y(),
+                     texelSize.z(),
+                     byteSize,
+                     channels,
+                     static_cast<Uint>(minValue),
+                     static_cast<Uint>(maxValue),
+                     avg,
+                     nonZero,
+                     avgNeighborDiff,
+                     byteSize > 0 ? static_cast<Uint>(bytes[0]) : 0u,
+                     byteSize > 1 ? static_cast<Uint>(bytes[1]) : 0u,
+                     byteSize > 2 ? static_cast<Uint>(bytes[2]) : 0u,
+                     byteSize > 3 ? static_cast<Uint>(bytes[3]) : 0u);
+    }
+
     static VkComponentSwizzle ToVkComponentSwizzle(TextureSwizzleParam swizzle) {
         switch (swizzle) {
         case TextureSwizzleParam::Red:
@@ -777,6 +873,8 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 break;
             }
         }
+        DumpTextureSyncStats(texture.GetExternalIndex(), texture.GetFormat(), uploadTarget, mipLevelCount,
+                             texelSize, byteSize, hasDirtyMipLevel);
         if (!hasDirtyMipLevel) {
             return true;
         }
@@ -1001,6 +1099,12 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             DeferViewRelease(resource.fullView);
             resource.fullView = VK_NULL_HANDLE;
         }
+        for (auto& sampledView : resource.perMipSampledViews) {
+            if (sampledView != VK_NULL_HANDLE) {
+                DeferViewRelease(sampledView);
+                sampledView = VK_NULL_HANDLE;
+            }
+        }
 
         const TextureFormatInfo formatInfo = ResolveTextureFormatInfo(texture.GetFormat());
         const VkComponentMapping sampledComponents = ResolveSampledViewComponents(texture, formatInfo);
@@ -1111,6 +1215,10 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 if (!uploadItems.back().expandedData.empty()) {
                     uploadItems.back().source = uploadItems.back().expandedData.data();
                 }
+                DumpTextureUploadStats(mipmapTexture.GetExternalIndex(), uploadItems.back().target,
+                                       uploadItems.back().level, uploadItems.back().texelSize,
+                                       uploadItems.back().source, uploadItems.back().uploadByteSize,
+                                       formatInfo.expandRgbToRgba ? 4u : 0u);
                 stagingSize += static_cast<VkDeviceSize>(uploadItems.back().uploadByteSize);
             }
         }
@@ -1156,16 +1264,17 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         VK_VERIFY(vkBeginCommandBuffer(commandBuffer, &beginInfo), "vkBeginCommandBuffer(texture)");
 
         const VkImageAspectFlags aspectMask = GetAspectMaskForFormat(outResource.format);
+        VkPipelineStageFlags uploadSrcStageMask = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
+        VkAccessFlags uploadSrcAccessMask = 0;
+        GetImageTransitionSourceState(outResource.layout, uploadSrcStageMask, uploadSrcAccessMask);
         Bool ok = TransitionImageLayout(commandBuffer, outResource.image,
-                              outResource.layout,
-                              VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-                               outResource.layout == VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL ?
-                                   kGraphicsSampledReadStages :
-                                   VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
-                               VK_PIPELINE_STAGE_TRANSFER_BIT,
-                               outResource.layout == VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL ? VK_ACCESS_SHADER_READ_BIT : 0,
-                               VK_ACCESS_TRANSFER_WRITE_BIT,
-                               aspectMask, 0, outResource.mipLevels, outResource.arrayLayers);
+                                        outResource.layout,
+                                        VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                                        uploadSrcStageMask,
+                                        VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                        uploadSrcAccessMask,
+                                        VK_ACCESS_TRANSFER_WRITE_BIT,
+                                        aspectMask, 0, outResource.mipLevels, outResource.arrayLayers);
         MOBILEGL_ASSERT(ok, "TransitionImageLayout to VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL failed");
 
         for (const auto& item : uploadItems) {
diff --git a/MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp b/MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp
index aae87450..29455b7c 100644
--- a/MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp
+++ b/MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp
@@ -20,6 +20,9 @@
 #include "MG_Util/Converters/MGToVk/RenderStateEnumConverter.h"
 #include "MG_Util/Converters/MGToVk/TextureEnumConverter.h"
 #include "MG_Util/Metrics/TextureMetrics.h"
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
 #include <vulkan/vulkan_core.h>
 
 namespace MobileGL::MG_Backend::DirectVulkan {
@@ -376,6 +379,94 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return attributeMask;
     }
 
+    static Bool ShouldDumpVertexInputStats() {
+        static const Bool enabled = [] {
+            const char* value = std::getenv("MOBILEGL_VERTEX_INPUT_STATS");
+            return value != nullptr && value[0] != '\0' && std::strcmp(value, "0") != 0;
+        }();
+        return enabled;
+    }
+
+    static const char* PresentDumpPath() {
+        const char* value = std::getenv("MOBILEGL_PRESENT_DUMP_PATH");
+        return value != nullptr && value[0] != '\0' ? value : nullptr;
+    }
+
+    static void DumpVertexInputStats(Uint32 location, const MG_State::GLState::VertexAttribute& attr,
+                                     Uint32 firstVertex, Uint32 vertexCount) {
+        if (!ShouldDumpVertexInputStats() || attr.Type != DataType::Float32 || attr.Size <= 0 || attr.Size > 4) {
+            return;
+        }
+
+        const SizeT componentSize = sizeof(float);
+        const SizeT elementSize = componentSize * static_cast<SizeT>(attr.Size);
+        const SizeT stride = attr.Stride > 0 ? static_cast<SizeT>(attr.Stride) : elementSize;
+        const Uint8* base = nullptr;
+        SizeT available = 0;
+        if (attr.Buffer) {
+            const auto& data = attr.Buffer->GetDataReadOnly();
+            if (!data || attr.Offset >= data->size()) {
+                return;
+            }
+            base = data->data() + attr.Offset;
+            available = data->size() - attr.Offset;
+        } else {
+            base = reinterpret_cast<const Uint8*>(attr.Offset);
+            available = static_cast<SizeT>(firstVertex + vertexCount) * stride;
+        }
+        if (base == nullptr || vertexCount == 0 || available < elementSize) {
+            return;
+        }
+
+        float minValues[4] = {0.0f, 0.0f, 0.0f, 0.0f};
+        float maxValues[4] = {0.0f, 0.0f, 0.0f, 0.0f};
+        Bool initialized = false;
+        SizeT sampled = 0;
+        const Uint32 maxSamples = std::min<Uint32>(vertexCount, 256);
+        for (Uint32 sample = 0; sample < maxSamples; ++sample) {
+            const SizeT offset = static_cast<SizeT>(firstVertex + sample) * stride;
+            if (offset + elementSize > available) {
+                break;
+            }
+
+            const auto* values = reinterpret_cast<const float*>(base + offset);
+            for (Int component = 0; component < attr.Size; ++component) {
+                if (!initialized) {
+                    minValues[component] = values[component];
+                    maxValues[component] = values[component];
+                } else {
+                    minValues[component] = std::min(minValues[component], values[component]);
+                    maxValues[component] = std::max(maxValues[component], values[component]);
+                }
+            }
+            initialized = true;
+            ++sampled;
+        }
+
+        if (!initialized) {
+            return;
+        }
+
+        std::fprintf(stderr,
+                     "MOBILEGL_VERTEX_INPUT_STATS loc=%u buffer=%u size=%d stride=%zu first=%u count=%u sampled=%zu "
+                     "min=(%.3f,%.3f,%.3f,%.3f) max=(%.3f,%.3f,%.3f,%.3f)\n",
+                     location,
+                     attr.Buffer ? attr.Buffer->GetExternalIndex() : 0u,
+                     attr.Size,
+                     stride,
+                     firstVertex,
+                     vertexCount,
+                     sampled,
+                     minValues[0],
+                     minValues[1],
+                     minValues[2],
+                     minValues[3],
+                     maxValues[0],
+                     maxValues[1],
+                     maxValues[2],
+                     maxValues[3]);
+    }
+
     static Bool TryGetCurrentVertexAttributeFormat(GLenum glType, VkFormat& outFormat) {
         switch (glType) {
         case GL_FLOAT:
@@ -1398,12 +1489,17 @@ void main() {
             if (binding >= vertexInputState.bindings.size()) {
                 break;
             }
+            const Uint32 bindingLocation = binding < vertexInputState.bindingAttributeLocations.size()
+                                               ? vertexInputState.bindingAttributeLocations[binding]
+                                               : static_cast<Uint32>(MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS);
+            if (bindingLocation < MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS) {
+                DumpVertexInputStats(bindingLocation, vao.GetAttribute(bindingLocation),
+                                     drawParams.firstVertex, drawParams.vertexCount);
+            }
             const Bool usesClientMemory = binding < vertexInputState.bindingUsesClientMemory.size() &&
                                           vertexInputState.bindingUsesClientMemory[binding];
             if (usesClientMemory) {
-                const Uint32 location = binding < vertexInputState.bindingAttributeLocations.size()
-                                            ? vertexInputState.bindingAttributeLocations[binding]
-                                            : static_cast<Uint32>(MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS);
+                const Uint32 location = bindingLocation;
                 MOBILEGL_ASSERT(location < MG_State::GLState::VertexArrayObject::MAX_VERTEX_ATTRIBS,
                                 "UploadAndBindVertexStreams failed to resolve client attribute location");
 
@@ -3840,7 +3936,8 @@ void main() {
         VkBufferObject presentStatsReadback;
         VkDeviceSize presentStatsReadbackSize = 0;
         const VkExtent2D presentStatsExtent = m_swapchainObject.GetExtent();
-        const Bool collectPresentStats = PresentStatsEnabled() && frame.isCommandRecording &&
+        const char* presentDumpPath = PresentDumpPath();
+        const Bool collectPresentStats = (PresentStatsEnabled() || presentDumpPath != nullptr) && frame.isCommandRecording &&
                                          presentStatsExtent.width > 0 && presentStatsExtent.height > 0;
         if (collectPresentStats) {
             presentStatsReadbackSize = static_cast<VkDeviceSize>(presentStatsExtent.width) *
@@ -3909,6 +4006,7 @@ void main() {
             MOBILEGL_ASSERT(pixels != nullptr, "Present stats: failed to map readback buffer");
             SizeT nonBlack = 0;
             SizeT nonTransparent = 0;
+            SizeT colored = 0;
             const SizeT pixelCount = static_cast<SizeT>(presentStatsExtent.width) *
                                      static_cast<SizeT>(presentStatsExtent.height);
             for (SizeT i = 0; i < pixelCount; ++i) {
@@ -3916,14 +4014,33 @@ void main() {
                 if (p[0] != 0 || p[1] != 0 || p[2] != 0) {
                     ++nonBlack;
                 }
+                const Uint8 minRgb = std::min(p[0], std::min(p[1], p[2]));
+                const Uint8 maxRgb = std::max(p[0], std::max(p[1], p[2]));
+                if (maxRgb - minRgb > 24) {
+                    ++colored;
+                }
                 if (p[3] != 0) {
                     ++nonTransparent;
                 }
             }
-            std::fprintf(stderr,
-                         "MOBILEGL_PRESENT_STATS nonBlack=%zu/%zu alpha=%zu/%zu size=%ux%u\n",
-                         nonBlack, pixelCount, nonTransparent, pixelCount,
-                         presentStatsExtent.width, presentStatsExtent.height);
+            if (PresentStatsEnabled()) {
+                std::fprintf(stderr,
+                             "MOBILEGL_PRESENT_STATS nonBlack=%zu/%zu colored=%zu/%zu alpha=%zu/%zu size=%ux%u\n",
+                             nonBlack, pixelCount, colored, pixelCount, nonTransparent, pixelCount,
+                             presentStatsExtent.width, presentStatsExtent.height);
+            }
+            if (presentDumpPath != nullptr) {
+                FILE* dump = std::fopen(presentDumpPath, "wb");
+                if (dump != nullptr) {
+                    std::fprintf(dump, "P6\n%u %u\n255\n", presentStatsExtent.width, presentStatsExtent.height);
+                    for (SizeT i = 0; i < pixelCount; ++i) {
+                        const Uint8* p = pixels + i * 4;
+                        const Uint8 rgb[3] = {p[0], p[1], p[2]};
+                        std::fwrite(rgb, 1, sizeof(rgb), dump);
+                    }
+                    std::fclose(dump);
+                }
+            }
         }
         frame.isCommandRecording = false;
         frame.hasCommandBufferRecorded = false;